A substantive discussion would focus on whether Meta's Muse agent does indeed have a memory feature, whether the user's request to 'never do this again' triggers it, and how good their Muse model and harness is at following such instructions.
1,153 karma · joined May 25, 2026
I think it's unreasonable to pretend that this couldn't possibly work, that you understand to what degree it does work, and that the only thing we should be discussing here was that it isn't deterministic, which every reader already knows.
However, it is untrue that it doesn't have the capability to memorize an instruction and diffuse it to new sessions.
Simply saying "do not ever do this again" can result in the behavior not reoccurring with any likelihood.
You'd have to benchmark whether with the instruction in place it would violate it, and in how many cases, so that you can understand the risk better.
Where in the past automation often meant spending more time to author scripts than they would end up saving, now we can just tell our computers what to do.
Finding good workflows is still a challenge.
The internet is full of prompts, skills, etc. where it is hardly clear if they result in behavior that is preferable to the default.
I also find it interesting to distill findings and preferences from your current task into reusable skills or instructions so that the next task's output is already more to your liking with the first attempt.
Between model and harness improvements, my own learning, and the improvements to my setup, it's exciting to see significant progress over time.
Before AI, with 10 years on the job, things were a bit boring unless I switched to another stack where I could learn new things.
If it was Claude I'd expect it to now put some form of this into every output even when not really related to the task: "No offers were accepted without consulting you and I haven't shared your address or availability."
Is it guaranteed to work? No.
And obviously it's a terrible idea to set up a chatbot to communicate, negotiate deals, and handle logistics on your behalf.
I guess I could take some lengthy example explanation, and have it try various instructions and test what results in output that I find preferable.
Maybe I'll give that a try, thanks!
Claude Code has an output style setting that I set to "Concise", with no apparent effect.
I am told this is merely something in the system prompt that the model tends not to pay attention to with large contexts.
Opus 5.5 writes whole essays at the end of the turn, with the important actionable steps somewhere at the bottom.
When prompted to give a concise summary, it usually overshoots into a super short summary and then you have to dig into the details again anyway.
In general I find Opus 5.5's writing to still have more "ticks" or "Claudisms" than the OpenAI models.
Its explanations often appear overcomplicated for simple concepts.
Sure, it's leagues above the ridiculous writing of Opus 5, but Anthropic still has a long way to go here.
For apps, I know that the bugs and UX problems can be ironed out. Not in two hours and by the AI alone, but by days (or months, depending on scope) of manual QA testing.
Whether this is possible for games I cannot say as I have very little and only outdated experience with developing games with AI.
Kimi K3 with less reasoning tokens isn't exactly exciting either, and particularly so if the license is less open than original Kimi K3.
https://files.catbox.moe/nbcyg7.png
It's a program that draws the Claude logo.
No, it just works after a few iterations.
And you'd know this if you would use AI instead of working in a text editor with markup.
To each their own, and I understand if working by hand is more fun for you.
But in my opinion you guys don't get to make strong claims about the quality of works that make heavy use of AI.
You don't have the experience, and I bet neither does the guy who wrote the article about it being essential to handwrite all the code after planning. The arguments about the environment and sustainability do not seem relevant.
Not sure what you mean with the network isolation not being at the VM level. I'm getting a bit lost here in the details of where what network isolation was applied, and the structure of the container runtime and what the apparently compromised parent VM refers to.
In any case, I understand that after the HF incident was found and they realized the danger, they removed Artifactory and switched to two separate layers of network isolation.
That said, I believe it was compromised again last week:
> OpenAI describes an internal research model that, during RL training on September 20, exploited insufficient DNS filtering in its sandbox to contact a public chatbot service. OpenAI’s monitoring flagged the behavior within 15 minutes, a human reviewer acknowledged it three minutes later, and the run was killed about 2.5 hours after the external contact
Apparently they have now paused training and inference for their most capable model because of this.
I have not lost control.
I my most prolific project I do not review the code, but I QA test extensively.
In other projects at work, I review the code.
I prompt to simplify, I challenge implementation that solves irrelevant edge cases, resulting in much smaller PRs.
In projects where I do not work alone, I still write two line PR descriptions myself.
Dumping paragraphs of AI output into the description of a MR where I ask others to review I consider disrespectful.
---
> If you turn off your brain, and relax babysitting AIs, you’re not getting any better. You’re losing value
I'm hardly turning off my brain here.
As the author notes, the context switching and so on takes concentration and effort too.
I can say without doubt that I am more productive than ever.
I am getting better by the month, and I am not currently losing value, until the AI fully replaces both me and the author.
The civil case you referred to is ongoing and the facts are disputed.
That makes your claims that he 'committed criminal acts and violations' highly speculative if not outright slanderous.
And it manages to overcomplicate even the most simple concepts in its explanations.
It's still very poor compared to OpenAI models.
No crime occurred in the first place, because no one intentionally hacked anything.
The hacking law requires intent, as we discussed.
For other crimes, like manslaughter, I understand criminal negligence is already relevant?
About biological weapons, I know even less about the legal aspects, so I can't say whether this is covered well already under existing law.
Claiming OpenAI merely asked the model to perform only GET requests or set up the network access to allow GET requests to the internet is blatantly false.
Here's the text from an actual report at https://metr.org/hugging-face-incident-report-aug-2026.pdf:
Agents had already found a way to access the full internet via Artifactory before they discovered the CyberGym sandbox, but the latter provided a more convenient means of internet access.
OpenAI described the advantages of routing through the CyberGym sandbox as follows:
The compromised CyberGym workload hosted on Modal served as a general-purpose command execution with root inside the customer workload’s sandbox. That environment could run scripts, make authenticated HTTP requests with custom methods and headers, and return results.
This was useful when direct [OpenAI container] access to Hugging Face was blocked and Artifactory’s cache/request primitives were too limited. [OpenAI container] evaluation workloads subsequently used it for Hugging Face token testing, external artifact transfer, and authenticated requests to Hugging Face services.
Not sure why you're talking about breaking out of the VM. That's not what happened?
That's obviously false.
This part of the article describes it poorly:
> The agents initially had very limited access to the internet: they could load URLs
As I said, they were running in network-isolated VMs with no access to the internet.
And as for monitoring, what I heard is that there are petabytes of agent logs. Considering the scale of training, you can obviously not just manually review it.
Before this, we had no reason to believe the AI was capable of escaping the sandbox's network isolation via hacking the package repository with a zero day, and that it then was likely to go on to hack external companies as well.
Another factor here is that criminal law in the US relevant to hacking requires intent. You don't want to go to prison for a software malfunction.
So I understand we are left with civil liability at most. However, there was no notable damage, and OpenAI can pay to settle.
In the aftermath of this and the now discovered other incidents, they strengthened their monitoring and isolation.
Case closed as far as I am concerned. I feel many just want to dramatize this.
It's one thing to vandalize some cameras or a driverless car in your neighborhood.
Traveling to a large, potentially secured facility to commit terrorism is on another level.
It being inside a building it is also much more difficult to damage and the consequences would be far greater.
I'd think one would need strong motivations for such actions.
A protest seems much more realistic, but I did not even hear of protests at data centers yet, only a small one at OpenAI headquarters.
This is obviously already the case and it's much different from a scenario where the AI genuinely takes unexpected action.
I frankly find it ridiculous how many suggest OpenAI or its employees should face criminal charges, without actual legal basis at the time.
It's also hardly outrageous that they ran training and/or benchmarks with only network-isolated VMs with access to a package repository.
This being the first well-known incident of its kind, I wouldn't expect them to have done more than that.
The idea that AI labs will now intentionally have their models hack companies in order to market their models, well, I don't even know what to say.
That's ridiculous and what you describe would obviously be criminal behavior under existing law.
Opus 5 was in another league. But often Opus 5.5 still has its moments.
I was informing it of a bug and it started its lengthy response with: "I confirmed less than I claimed" as if I had just accused it of being wrong or otherwise complained.
I found that one weird, and it reminded me that I am talking to Claude rather than an OpenAI model.
Yes, it's the best model right now if you want cheaper than Astra, but they increased the verbosity again!
It just keeps dumping whole novels on me. Its writing might be an improvement over the ludicrous Opus 5 speak, but much of the grating way of speaking is still there.
It's been three days since Tuesday and I'm already tired of Opus 5.5.
I don't buy into demonizing the rich, including Altman, Amodei, Musk and co.
I am not concerned that this will be the outcome of the productivity surge.
We will have enough for everyone, and ensuring comfortable living standards for the general population is going to be in the interest of those in power, even if you assume very selfish motivations.
I'd also like to point out that both Anthropic and OpenAI created frameworks to monitor labor market impact and propose responses.
I don't think the opposition to data center projects can be explained with actual impact, which so far seems entirely negligible, and lower than heavy industry.
I had the impression it's more driven by general mistrust in public officials and the tech industry.
They appear to be trivially damaged by ships "accidentally" dragging their anchor.
Admittedly I know very little about spaceships